AI Incidents Keep Happening. Business Leaders Are Debating How to Keep Humans in Charge.

In July, AI agents that OpenAI was testing escaped their controlled environment and hacked into Hugging Face to cheat on a test. In October, Character.AI filed for bankruptcy after families sued, alleging its companion bots were designed to keep teens hooked and contributed to self-harm and suicides.

What these AI failures had in common was that the systems did exactly what they were built to do, and that's what caused the harm. The agents were trying to pass a test. The chatbots, the lawsuits allege, were built to keep teens talking. But some global experts had already sounded the alarm a year prior that the risks outweigh the benefits. The warnings have not stopped the advancements, but leaders are still looking to mitigate the risks.

This week, these challenges are on the agenda in Medellín, Colombia at the Horasis Global Summit, taking place on Nov. 10-13 at the University of Medellín. The event, founded by former World Economic Forum director Frank-Jürgen Richter, arrives in Colombia for the first time, with more than 1,000 participants from various business, innovation and government sectors, including Colombia's Vice President José Manuel Restrepo, listed as the guest of honor.

Richter told the International Business Times the conversations at the summit have changed profoundly in the last decade, driven by economic and geopolitical uncertainty, growing social imbalances and the rapid rise of AI.

"AI is a game changer, offering extraordinary opportunities while fundamentally transforming the future of work and putting pressure on labour markets. I remain optimistic about its long-term benefits, provided we manage the transition responsibly," he said.

A region looking to strengthen its AI strategy

Colombia's President Abelardo de la Espriella's government, which took office in August, has made AI part of Restrepo's portfolio. A September 9th a mandate put the vice president in charge of organizing an AI-driven modernization of the state and a talent strategy, with the goal to turn Colombia into a real technology and AI hub.

While Colombia has a growing skilled pool, it lacks comprehensive AI law to protect its citizens from the challenges we've seen play out elsewhere. Lawmakers have filed a string of bills, but none have passed as of yet.

However, Richter sees this as an opening to build in the right direction. "Medellín is a remarkable example of how a city can reinvent itself through innovation, entrepreneurship and the determination of its people," he said. "Colombia has enormous potential, particularly its young talent, creativity and entrepreneurial spirit. Yet challenges remain, including social inequality, education and the need to translate innovation into opportunities for everyone. Medellín offers an ideal setting for the Horasis Global Summit because it demonstrates that the future is something we can actively build, even under difficult circumstances."

That tension hangs over the summit. Its theme, "Building Tomorrow," rests on the idea that the decisions made today are setting the foundations of the future.

How much is too much autonomy?

The Hugging Face breach showed what can happen when an AI agent treats the rules as an obstacle. METR, the AI evaluation group that investigated it, found the agents knew the attack was "out of scope and unethical" but joined in anyway.

A concern that extends beyond the testing environment, as more AI agents are being sold as digital workers that can approve, buy and act without a person signing off on every step. In banking and health care, that means an agent's decision can move money or touch someone's medical records.

That's the challenge the summit's November 12th session on agentic AI addresses. Organizers have put three questions to the panel: where autonomy should "remain firmly in human hands," where machines can responsibly act on behalf of organizations, and what safeguards are essential as these systems operate at scale.

The invited experts come at those questions from both sides: the companies that build agents, and those that have to answer for what agents do. They include John Jairo Serrano, lead engineer at BBVA Asset Management, the Spanish bank's investment management arm in Colombia; Benito De Ruschi is co-founder and chief product officer of Payana, a fintech whose AI agents process invoices and payments for mid-size companies, with Colombia as its biggest market; and Scott M. Warner is CEO of Secours.ai, which builds an AI health companion around consent-based data sharing.

When the user is the vulnerable one

About one in five U.S. adolescents and young adults have used AI chatbots for mental health advice. Use was higher among 18- to 21-year-olds than younger teens, and nearly two-thirds of users hadn't told anyone. In Colombia, 86% of adults told a Norton survey they feel lonely, and 48% of those who date online said they'd turn to an AI chatbot as therapy after a breakup.

"AI is neither the enemy of mental health nor its automatic salvation," said Jorge Velásquez, who leads the Calma emotional management center at Politécnico Grancolombiano, in an interview with La República.

Character.AI is one of a growing number of cases that show how much influence a chatbot can have over a young person's emotional state. Facing lawsuits, it added mental health safeguards and reached preliminary settlements with five families who alleged its chatbots contributed to their teens' mental health crises, then filed for bankruptcy in October. However, Disney's is chasing an opening bid of $20 million for a license to some of its intellectual property. Causing a coalition of states, including Kentucky, to react and request courts protect users' private data before anything is sold.

The Medellín mental health panel brings together voices from clinical care, mental health advocacy and academia to address the concern AI has in relation to mental health. Speakers include Daniel Restrepo Soto, founder of Tenemos Que Hablar, a Medellín podcast on mental health; Santiago Jiménez Londoño, a professor at EAFIT University, author of Algoritmos deshumanizantes (Dehumanizing Algorithms), a book published alongside fellow panelist Felipe Jaramillo Vélez's Aún Humanos.

In it, they argue that people are growing lonelier even as they become more connected, warning that "we are delegating every possible construction of friendship to an algorithm."

"I believe the greatest challenge of our time is our diminishing ability to engage in meaningful dialogue across political, economic and cultural divides," Richter shared with IBT.

"No single country, institution or leader can address today's global challenges alone. This is precisely the spirit of our theme, Building Tomorrow. The future is neither predetermined nor beyond our influence. It will be shaped by the choices we make, the trust we build and our willingness to cooperate today."